[CLOSED] Renal image registration evaluation

Status update (9 sept 2021)

This task force is now closed. 


Aim

Image registration is one of the components of renal MRI biomarkers that influence the final biomarker reliability. Renal image registration faces with specific difficulties and, consequently, it is not possible to reliably select optimal registration method based on results obtained for other registration tasks. There is no agreement on a renal registration quality measurement, which makes a quantitative comparison between registration algorithms difficult. In this task, we wish to define and implement strategies for specific renal image registration tasks to overcome this problem. Specifically, we wish to provide everything needed to test renal registration algorithms, including renal images, evaluation data and evaluation algorithm. This would enable users of renal MRI registration methods to evaluate their algorithms or select more reliable algorithms. This would also enable the image registration community to focus on the development of better renal registration algorithms and consequently contribute to the reliability of renal MRI biomarkers.

Steps identified to reach the goal and current status:

  • Selection of input data set(s).            
    - Open. Searching for datasets that could be publicly shared.
  • Preparation of the evaluation plan, considering multiple evaluation criteria base don segmented feature points, segmented regions, volume change, deformation extent, etc.
    - A decision was made to start with the feature-point based evaluation, which could later be supplemented with other criteria.
  • Segmentation - of feature points or regions, which includes selection of a segmentation tool and manual segmentation.
    - Most radiotherapy planning tools enable segmentation of points (supported in DICOM RT). A tool for segmenting feature points in Matlab is under development. 
  • Preparation of an evaluation algorithm.
    - The initial version of point-based evaluation algorithm in Matlab was developed and is currently being tested on other types of data.
  • Performing registration - users and developers of registration algorithms would register the given data and contribute their results.
  • Performing evaluation and publishing results.
  • Task promotion - attract multiple groups with different registration algorithms.
  • Write and publish a paper with all the task details. 

The evaluation task was first discussed on meeting in Koper, 3-4.3.2020. The decisions reached are evident from the meeting minutes.


Invitation

You are kindly invited to contribute to this task. There are many ways  one could contribute, including by sharing data, performing segmentation, defining the evaluation criteria, developing evaluation software. Your ideas are also welcome. Please email peter.rogelj@upr.si to join the task force.


Last updated: 30.1.2021